<p>Pore defect in laser powder bed fusion is governed by coupled process parameters and intermediate pore-structure, but most existing methods rely on direct process-response mapping and provide limited pathway interpretability. This study develops a hierarchical structural equation model for analyzing and predicting pore defects in LPBF 316L stainless steel. The proposed framework links laser power, scan speed, hatch spacing, and layer thickness with image-derived latent constructs, including pore morphology, edge characteristics, spatial distribution, network connectivity, size complexity, and size heterogeneity, before predicting final porosity. The model was established using 1040 image records from 65 process-parameter sets. The measurement model showed acceptable reliability, with composite reliability values of 0.76–0.90 and AVE values of 0.53–0.75. Path analysis confirmed that porosity is transmitted through a staged process-structure-porosity pathway, with strong effects from network connectivity to size complexity (0.74), size complexity to size heterogeneity (0.56), and network connectivity to porosity (0.58). Compared with MLR, GB, and RF, H-SEM achieved the best predictive performance, with R² = 0.84, RMSE = 2.22%, and MAE = 1.09%. This work provides an interpretable model for porosity prediction in LPBF, and supports the development of interpretable models for defect analysis in additive manufacturing.</p>